Dataset Imports (targetcl. low risk rating)

Basic characteristics Imports (targetcl. low risk rating)

71

target objects

The Automobile Database from UCI, containing several nominal values. The nominal values are replaced by natural numbers, artificially creating an ordering in the feature. The numbers are assigned according to the alphabetic order of the ordinal labels. The two classes are defined as those cars that have a assigned insurance risk rating (first feature) larger or smaller (and equal) than 0 respectively. The target class is low risk.Entries with missing values have been removed. Download mat-file with Prtools dataset.

88

outlier objects

25

features

Unsupervised PCA Imports (targetcl. low risk rating)

On the left, the PCA scatterplot is shown, on the right the retained variance for varying number of features.
On the left, the PCA scatterplot is shown of data rescaled to unit variance, on the right the retained variance.

Supervised Fisher Imports (targetcl. low risk rating)

On the left, the Fisher scatterplot is shown, on the right the ROC curve along this direction.

Results Imports (targetcl. low risk rating)

The experiments are performed using dd_tools. A rudimentary explanation of the classifiers is given in the classifier section.

513, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 73.6 (0.7) 73.7 (0.8) 51.2 (1.6)
Min.Cov.Determinant NaN (0.0) NaN (0.0) 48.5 (2.4)
Mixture of Gaussians 77.2 (1.5) 79.2 (0.9) 64.8 (2.5)
Naive Parzen 76.5 (1.5) 76.4 (1.4) 58.0 (2.5)
Parzen 87.5 (1.0) 87.3 (0.9) 82.1 (1.3)
k-means 65.1 (2.4) 75.3 (1.4) 57.0 (2.4)
1-Nearest Neighbors 87.6 (1.0) 87.3 (1.0) 82.1 (1.4)
k-Nearest Neighbors 87.6 (1.0) 87.3 (1.0) 82.1 (1.4)
Nearest-neighbor dist 84.5 (2.6) 85.0 (2.6) 72.0 (3.4)
Principal comp. 73.1 (0.7) 64.6 (1.8) 51.0 (3.0)
Self-Organ. Map 78.8 (1.0) 81.6 (2.3) 72.3 (2.1)
Auto-enc network 68.7 (3.0) 76.2 (1.9) 53.2 (1.5)
MST 87.1 (1.0) 87.5 (1.0) 82.7 (1.2)
L_1-ball 50.5 (31.5) 50.5 (31.5) 50.4 (2.4)
k-center 74.3 (2.6) 78.3 (3.1) 67.2 (3.9)
Support vector DD 86.8 (1.4) 87.0 (1.0) 81.9 (1.4)
Minimax Prob. DD 87.1 (0.9) 87.4 (1.1) 82.3 (1.3)
LinProg DD 78.5 (0.9) 83.5 (1.0) 73.0 (1.4)
Lof DD 81.0 (1.1) 81.2 (1.5) 78.8 (1.9)
Lof range DD 77.5 (1.6) 82.6 (2.1) 76.2 (2.3)
Loci DD 77.6 (1.0) 82.0 (0.5) 72.1 (1.4)

Classifier projection spaces The first classifier projection spaces are obtained by computing the classifier label disagreements (setting the threshold on 10% target error) and applying an MDS on the resulting distance matrix between classifiers:



Original



Unit variance



PCA mapped

Classifier projection spaces The second versions of the classifier projection spaces are obtained by computing the classifier ranking disagreements and applying an MDS on the resulting distance matrix between classifiers:



Original



Unit variance



PCA mapped